Imported from archit15singh/oss-contribution-strategy (AGENTS.md). Install upstream with npx skills add archit15singh/oss-contribution-strategy. Copyright stays with the author.

OSS Contribution Strategy

Context

Senior backend/AI engineer. Background at Abnormal AI (detection engineering, security). Targeting roles at OpenAI, Anthropic, Databricks, Cursor (or any top-tier AI infrastructure company). Goal: maximize resume ROI from open source contributions.

Core Strategy

• Primary: SigmaHQ — asymmetric advantage from detection engineering experience. Fastest path to meaningful merged PRs. Build reputation first.
• Secondary: vLLM — long-term play for AI infrastructure credibility. Pick unlabelled bugs, not crowded "good first issues".
• Drop: Qdrant, OpenTelemetry, LangChain, FastAPI (too crowded, too slow, or no edge)

Key Insight

Depth over breadth. 1-2 projects done deeply > surface-level PRs across 6.

Maintainer landscape (Jul 2026)

• Low-quality AI-generated PRs overwhelming maintainers
• High-value: well-tested fixes, thoughtful design discussions, long-term maintenance
• Source: ft.com (vibe-coding cleanup article), Times of India (GH India report)

Priority Queue

SigmaHQ

Why: Detection engineering asymmetric advantage. Issues are fresh (rule proposals, FP reports). Fast merge velocity. No competition from AI-generated PRs (requires domain knowledge).

Entry points:

• Rule proposals from open issues
• False positive reports from open issues
• Coverage gaps (new exploits, TTPs)

Goal: 3-5 merged PRs in first 30 days.

vLLM

Why: Hottest LLM serving project. High signal for AI infra roles.

Entry points:

• Unlabelled bugs (avoid crowded "good first issues")
• Python-level refactors
• Documentation / testing improvements

Goal: 1 meaningful PR in first 60 days.

Repos tracked

Commands

# Check SigmaHQ open issues
gh issue list -R SigmaHQ/sigma --limit 20 --json number,title,createdAt,labels,url

# Check SigmaHQ recent merged PRs
gh pr list -R SigmaHQ/sigma --state merged --limit 10 --json number,title,mergedAt,author

# Check vLLM open issues (unlabelled)
gh issue list -R vllm-project/vllm --limit 30 --json number,title,labels,createdAt

# Check vLLM help wanted (not good first issue — less crowded)
gh issue list -R vllm-project/vllm --label "help wanted" -L 10 --json number,title,createdAt,comments

Gotchas

• SigmaHQ issues are mostly rule proposals and FP reports — domain work, not code
• vLLM "good first issues" are instantly flooded with volunteers (10+ per ticket)
• Qdrant good-first-issues are stale/blocked or taken
• OpenTelemetry "good first issues" are years old with spec baggage
• Microsoft Founders Hub no longer gives direct OpenAI credits (discontinued mid-2025) — only Azure OpenAI Service
• OpenAI Codex for Open Source program exists alongside Claude for Open Source — apply to both
• Codex Open Source Fund ($25K API credits) is a SEPARATE application from Codex for OSS (ChatGPT Pro)
• Databricks for Startups relaunched June 2026 — consolidated into one program, now includes Neon (Postgres)
• Do NOT use utm_source query params when linking to repos